estimating platform market power in two-sided markets with an
TRANSCRIPT
Estimating Platform Market Power in Two-sided Marketswith an Application to Magazine Advertising
Minjae Song
1Simon Graduate School of BusinessUniversity of Rochester
2011 FTC Micro Conference
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 1 / 21
In Two-sided Markets
Two groups of agents interact through a platform.
Each group cares about the presence of agents on the other side, andthus the decisions of agents on one side affect the utility of agents onthe other side.
Platforms account for these cross-group externalities in makingstrategic decisions (e.g. setting prices).
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 2 / 21
Examples
Payment systems
Merchants and consumers interact through credit cards.
Video game systems
Game developers and game players interact through video consoles.
Advertising in newspapers/magazines/websites
Advertisers and readers interact through media platforms.
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 3 / 21
What I do in this paper
My paper brings two important features of the two-sided market intoa structural model.
Agents on each side care about the presence of agents on the otherside.Platforms charge two prices, one for each group.
I focus on cases where platforms charge fixed membership fees.
I consider two versions of the two-sided market.
Two-sided single homing: agents on both sides join one platform each.Competitive bottleneck: agents on one side join one platform butagents on the other side join multiple platforms.
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 4 / 21
I show how to estimate agents’demand (preferences) for platformsusing data on (two) membership prices, the number of agents onplatforms, and other platform attributes.
The presence of agents from the other side is an important platformattribute and this variable is an endogenous variable.
Given demand estimates, one can recover platforms’costs of servingagents and measure their markups (market power).
Price elasticity does not have a closed form because of the so-calledfeedback loop effect.There are two demand equations, one for each group, and both shouldbe used simultaneously to recover the costs.
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 5 / 21
Literature
Numerous theory papers on two-sided markets.
The most cited ones are Rochet and Tirole (JEEA 2003; RAND 2006)and Armstrong (RAND 2006).My paper is closely related to Armstrong (2006) .
Relatively few empirical papers but the number is growing fast.
Rysman (RESTUDS 2004) on the Yellow Page market - zero price forconsumers.Argentesi and Filistrucchi (JAE 2007) on the Italian newspaper market- consumers do not care about advertising.
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 6 / 21
Model 1: Two-sided single-homing
Two groups of agents, groups A and B. Each group cares about thepresence of the other group on platforms.
There are J platforms competing to attract agents from both sides.If platform j attracts sAj and s
Bj portions of the two groups, agents’
utilities are
uAij = µAj + αAsBj − λApAj + ξAj + εAij
uBij = µBj + αB sAj − λBpBj + ξBj + εBij
Consumers may choose the outside option of joining no platform andreceive zero mean utilities and an idiosyncratic shock.
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 7 / 21
Assuming εij is distributed the type I extreme value, platform j ′smarket shares are
SAj(pA , sB , ξA |Ω
)=
exp(
µAj + αAsBj − λApAj + ξAj
)1+∑Jm=1 exp
(µAm + αAsBm − λApAm + ξAm
)SBj(pA , sB , ξA |Ω
)=
exp(
µBj + αB sAj − λBpBj + ξBj
)1+∑Jm=1 exp
(µBm + αB sAm − λBpBm + ξBm
)
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 8 / 21
Model 2: Competitive bottleneck
In the competitive bottleneck model, while one group, say group A,deals with a single platform (single-homes), the other group, saygroup B, wishes to deal with multiple platforms (multi-homes).
A good example is media advertising.
For group A agents I use the same utility function used in thesingle-homing model except that I use the number of group B agentsinstead of the share.
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 9 / 21
I follow Armstrong (2006) to model group B agents’membershipdecision. I assume that she makes a decision to join one platformindependently from her decision to join another. She joins a platformas long as its net benefit is positive.Given the fixed membership fee, say pBj , a type-α
Bi agent will join
platform j ifαBi ωjn
Aj ≥ pBj .
Suppose platforms only know the distribution of αBi . Since each groupB agent is ex ante identical, a platform will charge a single price pBjand the number of group B agents joining platform j is determined by
SBj(pB , sA |Ω
)=
(1− F
(pBj
ωjnAj|θ))
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 10 / 21
Computing price elasticities
Because of the cross-group externalities
∂SAj(pA, sB , ξA |Ω
)∂pAk
6=∂sAj∂pAk
This makes elasticity computation an implicit function problem.Treating share equations as an implicit function, the elasticity can becomputed using the Implicit Function Theorem.For example, in the competitive bottleneck model,
FAj (s,p) ≡exp
(µAj + αAsBj M
B − λApAj + ξAj
)1+∑Jm=1 exp
(µAm + αAsBmMB − λApAm + ξAm
) − sAj = 0FBj (s,p) ≡
(1− G
(pBj
ωj sAj MA|θ))− sBj = 0
for j = 1, ..., J. where s are endogenous variables and p are controlvariables.
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 11 / 21
Estimation: Two-sided Single-home Model
With observed market shares treated as one of equilibria, I estimatethe following system of equations
log(sAj)− log
(sA0)
= µAj + αAsBj − λApAj + ξAj
log(sBj)− log
(sB0)
= µBj + αB sAj − λBpBj + ξBj
j = 1, ..., J .The model parameters are Ω =(
µAj , µBj ,λ
A,λB , αA, αB).
The demand-side model can be consistently estimated by the GMMwith IVs.
In addition to the price variable, the other group’s share variable is alsoan endogenous variable.
This variables is correlated with(
ξAj , ξBj
)for all js because of the
feedback loop.
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 12 / 21
Estimation: Competitive Bottleneck Model
For group A agents we have the following equation to estimate
log(sAj)− log
(sA0)= µAj + αAnBj − λApAj + ξAj
For group B agents ωj is recovered by inverting the second share
equation with a given value of θ and data on(nBj , n
Aj , p
Bj ,MB
).
Assuming that ωj is a function of platforms’non-price characteristics,we have another equation to estimate
ωjt = f(xjt |βB
).
where ωjt is computed by inverting
nBjt =
(1− F
(pBjt
ωjtnAjt|θ))
MB
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 13 / 21
Recovering marginal costs and markup
Demand estimates are used to recover platforms’costs using theprofit maximization condition. Assuming the constant marginal cost,platform j’s profit is
πj =(pAj − cAj
)sAj MA +
(pBj − cBj
)sBj MB
where MA and MB denote the total number of agents for each grouprespectively.The profit maximizing first order conditions are
∂πj
∂pAj= sAj MA +
(pAj − cAj
) ∂sAj∂pAj
MA +(pBj − cBj
) ∂sBj∂pAj
MB = 0
∂πj
∂pBj= sBj MB +
(pBj − cBj
) ∂sBj∂pBj
MB +(pAj − cAj
) ∂sAj∂pBj
MA = 0
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 14 / 21
The two marginal costs should be searched simultaneously. Thissearch process involves numerical computation of the own- andcross-price elasticities as derivatives of the implicit function for eachset of trial values.
Platform’s markup from one group is a function of its markup fromthe other group.
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 15 / 21
Empirical application
Advertising in magazines. Magazines serve readers on one side andadvertisers on the other side.
Panel data (1992 to 2010) on TV magazines in Germany.
Quarterly information on copy prices, advertising rates, advertisingpages, content pages, and circulation are collected from a non-profitpublic institution equivalent to the US Audit Bureau of Circulation.
Finding IVs from different magazine segments (Kaiser and Song, IJIO2009).
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 16 / 21
Data
There are about 10 to 15 magazines in each quarter published by 5 to7 publishers.
Each copy is sold at around 1 Euro, while one page of advertising issold at around 30,000 Euros.
The average magazine sells about 1.5 million copies in each quarter,has about 1,000 content pages and about 250 advertising pages.
The average magazine’s revenue from selling copies is about 1.5million Euros, while its advertising revenue is 7 million Euros.
It is hard to argue that the copy price covers the publishing cost. 1Euro for an over 100 page magazine seems unreasonably low.However, the low copy price is not unreasonable in the two sidedmarket.
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 17 / 21
Estimation results
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 18 / 21
Magazine (Platform) markup
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 19 / 21
Merger Analysis
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 20 / 21
Summary
My structural model has two key features of the two-sided market.
Both groups care about the presence of the other group, so thecross-group externalities are present on both sides.Platforms set different prices for each group to maximize joint profitsfrom both sides.
The empirical results show that most magazines set copy prices belowmarginal costs to increase the reader basis and make profits fromselling advertising space.
When the advertising side is ignored, the same demand estimatesimply high markups on the reader side.
Counterfactual exercises show that platform mergers do notnecessarily increase copy prices and, as a result, readers may notnecessarily be worse off in more concentrated markets.
Song (University of Rochester) Two-sided Markets 2011 FTC Micro Conference 21 / 21